Early development
A home for your agents.
On your infrastructure.
Radhouse is a self-hosted platform being built for persistent AI agents: agents that retain their workspaces, work on useful assignments, and collaborate within permissions you control.
Where it actually is
Durable Hermes execution and the everyday operator workflow are implemented. Researcher has an owner-attested standard Buzz identity and private conversation. OIDC composition, administrator invitation/recovery UX, and a qualified fleet release remain incomplete.
This page is not a claim that Radhouse V1 is released. The project is in pre-release implementation and qualification.
What it is
Running an agent in a dedicated VM is a useful beginning. Managing several agents introduces more questions: who can use them, what they can access, how they share work, which tasks need attention, and how to keep their environments useful between assignments.
Radhouse aims to make that everyday experience obvious while giving the person running the infrastructure explicit, inspectable controls. The name combines RAD — Running Agents Digitally — with a welcoming home for your agents. Say RAD-house.
Who it's for
Tinkerers, self-hosting enthusiasts, homelab operators, and people who care about security. The people using assigned agents should not need infrastructure expertise — administrators manage infrastructure and access; operators work with assigned agents day to day.
Flexible by design. Opinionated by default.
- Persistent agents: a Linux VM for each agent, with retained files, tools, and workspace; Hermes as the first supported agent runtime.
- Two deployment paths: guided Proxmox provisioning and installation into suitably isolated Linux VMs supplied by the operator.
- An approachable control plane: administrator, operator, and viewer roles, with a simple operator work home for assigned agents and current work.
- Private and shared work: private agents by default, explicit sharing, disclosed supervision, and human-reviewed publication.
- Understandable boundaries: visible work context, audience, and bot access at the point of use.
- Security and maintenance oversight: scoped security findings, update checks, and automatic weekly OS updates with advance notice and work preservation.
These are release requirements. Runtime compatibility, isolation, recovery, network profiles, and shared-service integrations still need qualification.